Traditional RDBMS Wisdom is All Wrong -- In Three Acts "

!

The “Stonebraker Says” Webinar Series!

The first three acts:!1.  Why the elephants are toast and why main memory is the

answer for OLTP!–  Today!

2.  Fast Data with ACID - May 13"

3.  Why main memory is still the answer - June 10!–  When your data doesnʼt fit!–  When you are doing event processing"

Our Speaker!

Dr. Mike Stonebraker of MIT, co-founder of VoltDB!

Overview of Today!

•  Trends in the marketplace!–  Some data and fluffy marketing slides!

•  One size fits none!

•  Why Traditional RDBMS Wisdom is All Wrong!

Version 4.0 launched in January 2014!

Offices in Bedford, MA & Santa Clara, CA

Partners in USA, Europe, Asia, and

South America!

400"Customers

Mike Stonebraker"Founder, Advisor!

VoltDB - Background

Magic Quadrant “Operational Databases”

2013 - One of the Top Companies that Matter

Most in Data

PACE OF CHANGE!

1 billion connected places

50 billion connected things

PLACES

PEOPLE

THINGS

Source:  Ericsson  2013

Smart Farm!

Digital

Physical

The Potential: Smart Everything!

•  Physical and digital worlds collide!•  But nothing happens without making things Smart…!•  Requirements for Smart: Fast, Unlimited Throughput!

Smart products!Smart places!Smart networks!Smart services!Smart solutions!

(ci=es,  farms,  buildings…)

Trends!

•  Internet: people + content + things!–  Digital and physical worlds collide!

•  Data: Big, complex, high velocity!

•  Technology: distributed, virtual, cloud!

One Size Fits None “The elephants are toast”"

Stonebraker Says!

Traditional RDBMS: The Elephants!

•  Sell code lines that date from the 1970ʼs!•  Legacy code!•  Built for very different hardware configurations!•  And some cannot adapt to grids….!

•  That was designed for business data processing (OLTP)!•  Only market back then!•  Now warehouses, science, real time, embedded, ..!

Current DBMS Gold Standard!

•  Store fields in one record contiguously on disk!

•  Use B-tree indexing!

•  Use small (e.g., 4K) disk blocks; heavily encoded!

•  Align fields on byte or word boundaries!

•  Conventional (row-oriented) query optimizer and executor!

•  Write-ahead log!

•  Row-level dynamic locking!

!

Terminology -- “Row Store” !

Record 2"

Record 4"

Record 1"

Record 3"

E.g. DB2, Oracle, Sybase, SQLServer, "Postgres, MySQL, Netezza, Teradata,…""

At This Point, RDBMS is “long in the tooth”"There are at least 6 (non trivial) markets where a row store can be clobbered by a specialized architecture !

–  Warehouses (Vertica, Red Shift, Sybase IQ)!

–  OLTP (VoltDB, Hana, Hekaton)!

–  RDF (Vertica, et. al.)!

–  Text (Google, Yahoo, …)!

–  Scientific data (R, MatLab, SciDB)!

–  Streaming data (coming in part 3 )!

Definition of “Clobbered”!

•  A factor of 50 in performance!

Pictorially:!

OLTP!Data Warehouse!

Other apps!

DBMS! apps!

The DBMS Landscape – Performance Needs!

OLTP!Data Warehouse!

Other apps!

low!

high!

high!

high!

One Size Does Not Fit All -- Pictorially!

Open source!

Vertica, et. al.! VoltDB, etc.!

SciDB, etc. ! !

Elephants get only ! “the crevices”!

Stonebrakerʼs Prediction: Coming True!

•  The DBMS market will move over the next decade !

•  To specialized (market-specific) architectures!

•  And open source systems!

•  To the detriment of the elephants!!

Traditional RDBMS Wisdom is All Wrong for OLTP"

Stonebraker Says!

Reality Check #1 on OLTP Data Bases!

§  TP data base size grows at the rate transactions increase!§  1 Tbyte is a really big TP data base!§  1 Tbyte of main memory buyable for around $30K (or less)!

§  (say) 64 Gbytes per server in 16 servers!§  If your data doesnʼt fit in main memory now, then wait a

couple of years and it will…..!§  Facebook is an outlier!

Reality Check #2: Client Interactions!

•  ODBC/JDBC is way too heavy to go fast!•  Need a lightweight protocol!

–  Asynchronous!–  Connection pooling!–  Stored procedure interface!

•  Otherwise “the high pole in the tent” is the client interface!–  Hekaton problem!!

Reality Check #3: Main Memory Performance!

•  TPC-C CPU cycles!•  On the Shore DBMS prototype!•  “Elephants” should be similar (or worse)!

To Go Fast….!

•  Better search data structures!–  E.g., fractal trees, better B-trees, …!

•  Affect the 4% piece of the pie!–  Irrelevant!–  Except in things like bulk load!

To Go Fast….!

•  Have to get rid of ALL FOUR pieces of the pie!•  Consider Oracleʼs TimesTen!

–  Main memory DBMS (one slice gone)!–  Row level locking!–  WAL!–  Multi-threaded!

•  3 slices remain!!

Implications….!

Slices  that  remain Rela0ve  performance

4 1    (elephants)

3 1.3  (TimesTen)

2 2

1 4

0 25  (Obvious  goal  of  VoltDB)

Removing Slice #1: Buffer Pool!

•  Main memory DBMS!–  Suppose my data is too big?!–  Stay tuned for part 3!

•  Any disk-based system will lose!!–  All the NoSQL guys!–  All the elephants!

Removing Slice #2: Latches!

•  A DBMS has a LOT of shared data structures!–  B-trees!–  Lock table!–  Head of the WAL!–  Sorting buffers!–  Malloc-able memory!–  Admission control data structures!–  …!

Removing Slice #2: Latches!

•  In a multi-core world, you need latches to avoid corruption!•  In a many core world, this will be a worse disaster!

–  All the traditional concurrency control systems degrade after (at most 32-64) cores!

Solutions!

•  Single-threaded code!–  VoltDB!

•  Latch-free data structures!–  Lots of research in this area!!–  So far, focused mainly on indexing and concurrency control!

•  What about all the other shared data structures?!

So Far….!

•  Elephants must convert to !–  Main memory DBMS!–  Single-threaded or latch-free or …!

•  A complete rewrite!–  See Clayton Christianson “The Innovatorʼs Dilemma”!

•  But it gets worse…!

Next Time – Fast ACID!

•  Recovery strategy!–  Aries WAL is dead; long live transaction logging!

•  Replication strategy!–  Active-active is the answer!–  In contrast to the elephantʼs active-passive!

•  Concurrency control strategy!–  Determinism wins; nobody uses row level locking!

•  More headaches for the elephants!!

Questions?!

•  Use the chat window to type in your questions.!

•  Try VoltDB yourself:!

Ø Free trial of the Enterprise Edition:!•  www.voltdb.com/download!

Ø Open source version is available on github.com!

§  Register for the May 13th webinar!

THANK YOU!" !